Optimized Indoor Natural Ventilation Design Integrating Thermal Comfort and Architectural Aesthetics: A Multi-Objective Thermodynamic Simulation Study

This study takes the thermodynamics of non-equilibrium open systems as the core theoretical support, treating the indoor ventilation space as an irreversible evolution system with continuous momentum, heat, and moisture exchange across boundaries. The internal flow, heat transfer, and mass transfer processes within the system continuously generate energy dissipation and entropy increase effects. The finite volume method is adopted to carry out full-domain numerical discretization and solution, combined with the k–ω Shear-Stress Transport (SST) turbulence model to accurately characterize near-wall flow behavior, turbulence evolution laws, and the transport process of thermal and humidity gradients. A coupled solution strategy is employed to perform synchronous calculations of indoor flow field, temperature field, and humidity field, ensuring the stability and physical authenticity of numerical results under complex natural ventilation conditions. PINNs are introduced to achieve efficient acceleration of simulation computations, using fluid dynamics governing equations and thermodynamic conservation laws as physical constraints, and combining high-fidelity numerical simulation samples to complete model training, thereby realizing rapid parameterized prediction of multi-physical fields. This paper does not elaborate on well-established turbulence numerical simulation methods and basic thermodynamic theories, but focuses on systematically explaining the self-developed coupling mechanism between architectural form and thermodynamic boundaries, the architecture of physics-constrained surrogate models, and the entropy-generation-mechanism-based multi-objective optimization system, thereby establishing a complete research framework deeply integrating theoretical mechanisms, numerical simulation, and intelligent optimization.

2.1 Aesthetics–thermodynamics parametric coupling model

Based on Non-Uniform Rational B-Spline surface modeling technology, this study performs continuous parametric representation of key envelope interfaces and air-guiding components in indoor natural ventilation systems, selecting five categories of core control variables capable of balancing both visual formal order and airflow guidance functions. These include the window opening aspect ratio R_aw, the Gaussian curvature of overhanging roofs K, the vertical taper angle of air guide towers, the installation inclination angle of guide louvers $\theta$, and the fractal dimension of porous ventilation grilles D_f. The value ranges of each variable are strictly defined in accordance with classical architectural aesthetic proportion rules and the operating conditions of natural ventilation structures, respectively set as: $0.618 \leq R_{a w} \leq 1.618,-0.05$. $\mathrm{m}^{-2} \leq K \leq 0.05 \mathrm{~m}^{-2}, 0^{\circ} \leq \theta \leq 30^{\circ}, 15^{\circ} \leq \beta \leq 75^{\circ}, 1.2 \leq D_f \leq 1.8$. All aesthetic indicators are encoded in continuous numerical form, breaking through the limitation of discrete selection in traditional architectural form design, and enabling continuous gradual variation and refined regulation of architectural appearance.

Different from existing studies that focus only on post-evaluation of completed forms for aesthetic quantification, this study directly transforms geometric aesthetic parameters into boundary constraint conditions for ventilation flow, establishing a direct connection between architectural form and fluid transport processes. Through independently developed parametric linkage logic, multi-dimensional morphological parameters can be updated synchronously, geometric models reconstructed in real time, and simulation boundaries automatically matched, forming an integrated dynamic representation system. This representation transforms architectural aesthetic elements from passive evaluation indicators to active control variables, embedding them in the entire workflow of ventilation numerical calculation, providing a stable variable basis for constructing quantitative coupling relationships between form parameters and turbulence characteristics, wall flow, and entropy generation evolution, and offering necessary support for realizing synergistic regulation of thermodynamic dissipation under aesthetic form constraints. Figure 1 shows the schematic diagram of the bidirectional dynamic coupling model between architectural aesthetic parameters and thermodynamic computation boundaries.

Figure 1. Schematic diagram of the bidirectional dynamic coupling model between architectural aesthetic parameters and thermodynamic computation boundaries

To achieve active linkage between architectural aesthetic form and thermodynamic dissipation processes, this study independently develops a Python–OpenFOAM bidirectional data interaction interface, overcoming the unidirectional limitation of traditional parameter transmission, enabling real-time writing, analytical mapping, and dynamic correction of geometric aesthetic parameters into the CFD boundary condition module. This constructs a quantitative correlation system among aesthetic form–wall flow–irreversible dissipation, fundamentally solving the core problem of disconnection between aesthetics and thermodynamic boundaries in existing studies. The interface connects to OpenFOAM’s boundary condition solvers through custom function libraries, capturing instantaneously the influence of changes in aesthetic parameters on turbulence evolution, momentum transport, and heat–mass exchange, ensuring the timeliness and accuracy of coupling relationships, and providing technical support for precise regulation of subsequent thermodynamic dissipation.

Accurate regulation of inlet turbulence vorticity is key to reducing ventilation irreversible dissipation, and the window opening aspect ratio, as a core aesthetic parameter, when deviating from the golden ratio constant, significantly intensifies inlet flow distortion and vorticity dissipation. Based on data from 30 groups of orthogonal pre-experiments, the analytical coupling relationship between window opening aspect ratio and inlet turbulence vorticity intensity is fitted as:

$\zeta=\zeta_0 \cdot\left(1+\alpha \cdot / R_{a w}-\varphi /\right)$          (1)

where, ζ is the corrected inlet turbulence vorticity intensity, ζ₀ is the baseline vorticity intensity, $\varphi$ is the golden ratio constant, and $\alpha$ is an empirical coefficient. Using the least squares method, $\alpha$ = 0.82 is calibrated, with a determination coefficient R² = 0.96, indicating very high fitting accuracy of this coupling relationship. This equation acts directly on the turbulence source term correction in CFD solutions, transforming changes in aesthetic parameters into adjustments of vorticity generation intensity in real time, realizing proactive control of flow irreversibility by aesthetic design, differing from the limitation of existing studies that analyze the indirect influence of aesthetics on flow only through post-evaluation.

The realization of coupling between wall roughness and thermo-mass boundaries further improves the full-dimensional regulation of aesthetic parameters on thermodynamic processes. Changes in the vertical taper angle of air guide towers alter wall flow separation characteristics, thereby affecting near-wall viscous dissipation. Based on experimental data, the linear coupling relationship between equivalent wall roughness height and taper angle is obtained:

$k_s=k_{s 0} \cdot(1+0.02 \theta)$           (2)

where, k_s is the corrected equivalent wall roughness height, k_s0 is the baseline roughness height, and θ is the vertical taper angle of the air guide tower. Combined with the wall function model, k_s is further transformed into roughness Reynolds number, and the near-wall momentum diffusion coefficient and viscous dissipation source term are corrected to achieve precise control of momentum transport by aesthetic parameters. Meanwhile, the Gaussian curvature of overhanging roofs corrects the wall heat flux density distribution through a quadratic fitting equation, correlating the heat exchange area gradient with thermal boundary layer thickness; the fractal dimension of porous grilles directly regulates the moisture diffusion resistance in porous media domains, realizing aesthetic coupling of mass transport boundaries. The above coupling mechanisms work synergistically, fully covering momentum, energy, and mass thermodynamic boundaries, achieving full-dimensional, proactive regulation of thermodynamic dissipation processes by architectural aesthetic parameters, forming one of the core technological innovations of this study.

2.2 Thermodynamic-constrained physics-informed neural network surrogate model

This study constructs a fully connected deep neural network as the carrier for surrogate prediction of thermodynamic field quantities. The core innovation lies in breaking through the limitation of single-input dimensionality in traditional surrogate models by establishing a joint input system combining aesthetic features and spatial positions, thereby realizing accurate mapping from architectural form parameters to full-domain thermodynamic field quantities. The input layer contains an 8-dimensional feature vector, of which 5 dimensions constitute a feature vector formed by multi-dimensional architectural aesthetic parameters, and the remaining 3 dimensions correspond to indoor spatial coordinates. The two are jointly input to synchronously capture the coupled influence of differences in aesthetic parameters and variations in spatial position on thermodynamic field quantities, overcoming the defect of traditional models that can only perform single-dimensional prediction and fail to correlate form with field distribution. The hidden layers of the network are set to 5 layers, with each layer fixed at 128 neurons. This structure is determined through cross-validation optimization, ensuring both sufficient fitting capability for complex aesthetics–thermodynamics coupling relationships and effective avoidance of overfitting and gradient vanishing problems, thereby guaranteeing the generalization performance of the model. Figure 2 shows the network architecture.

To balance model training efficiency and the physical plausibility of prediction results, a layered activation strategy is adopted for precise regulation, which constitutes another core innovation of the surrogate model in this study. Shallow layers employ the ReLU activation function, expressed as:

$\operatorname{Re} L U(x)=\max (0, \mathrm{x})$          (3)

which effectively mitigates gradient saturation effects, accelerates convergence, and enhances the extraction capability for high-dimensional input features. Deep layers adopt the Sigmoid activation function, expressed as:

$\operatorname{Sigmoid}(x)=1 /\left(1+e^{-x}\right)$           (4)

which constrains output field quantities within reasonable physical intervals, avoiding non-physical solutions such as negative values in core thermodynamic parameters including entropy generation rate and exergy temperature, thereby ensuring the physical authenticity of prediction results. The output layer adopts a dual-output structure, synchronously predicting the total entropy generation rate field and the exergy temperature field, eliminating the need to construct separate independent surrogate models. This approach not only greatly improves computational efficiency but also avoids error accumulation caused by separate-model prediction, providing efficient and accurate field quantity data support for subsequent multi-objective optimization, and realizing seamless integration between the surrogate model and the optimization process.

Figure 2. Architecture of the thermodynamic Physics-Informed Neural Network (PINN) surrogate model embedded with multi-source physical constraints

To address the core shortcoming of traditional PINNs whose prediction results tend to deviate from the fundamental laws of thermodynamics and fluid mechanics, this study constructs a multi-source thermodynamic constraint loss function, deeply coupling data fitting accuracy with physical conservation laws, thereby achieving simultaneous improvement in prediction accuracy and physical plausibility. This constitutes a key innovation distinguishing the surrogate model of this study from existing research. The total loss function adopts a weighted coupling form, forcing the model to strictly follow the laws of incompressible fluid flow and the second law of non-equilibrium thermodynamics while learning the features of high-fidelity simulation data. Its expression is:

$L_{\text {total}}=L_{\text {data}}+\lambda_1 L_{N-S}+\lambda_2 L_{\text {ent}}$            (5)

The weighting coefficients λ₁ and λ₂ are optimized through ten-fold cross-validation, with values of 0.3 and 0.5 respectively. This weight allocation highlights the central role of entropy generation constraints in thermodynamic optimization while balancing the necessity of fluid mechanics constraints. The data fitting loss uses 200 sets of high-precision CFD simulation snapshots as ground truth, primarily constraining the fitting errors of the predicted entropy generation field and exergy temperature field, thereby ensuring the numerical accuracy of prediction results. The fluid mechanics constraint loss embeds the residuals of the incompressible Reynolds-Averaged Navier–Stokes equations and the energy conservation equation, effectively controlling the physical compatibility of velocity, pressure, and temperature gradients, and preventing non-physical solutions that violate fluid flow laws.

The core innovation lies in introducing an original entropy generation constraint term, filling the research gap in traditional surrogate models that lack precise constraints on thermodynamic irreversible dissipation. By decomposing three categories of core irreversible entropy generation terms in indoor ventilation processes, a complete entropy generation constraint system is constructed, achieving deep adaptation between the surrogate model and thermodynamic optimization objectives. The specific expressions of the three categories of irreversible entropy generation components are:

$\left\{\begin{array}{c}\dot{s}_{\text {gen,cond}}=\frac{k}{T^2}|\nabla T|^2 \\ \dot{s}_{\text {gen,visc}}=\frac{\mu \Phi}{T} \\ \dot{s}_{\text {gen,hum}}=\frac{R}{C} \frac{D \rho_v}{D t}\end{array}\right.$           (6)

where the thermal conduction entropy generation term originates from irreversible energy dissipation caused by temperature gradients, the viscous dissipation entropy generation term arises from fluid viscous friction, and the humidity diffusion entropy generation term quantifies the irreversible loss of mass transport during human metabolic heat and moisture release. Φ denotes the viscous dissipation function, k and μ represent the fluid thermal conductivity and dynamic viscosity, respectively. The total entropy generation rate is the sum of the three components, expressed as:

$\dot{S}_{\text {gen,tot}}=\dot{S}_{\text {gen,cond}}+\dot{S}_{\text {gen,visc}}+\dot{S}_{\text {gen,hum}}$           (7)

The entropy generation constraint term controls the conservation residuals of the total entropy generation rate and each component, forcing the entropy generation field predicted by the model to strictly satisfy non-equilibrium thermodynamic laws, eliminating unreasonable phenomena such as negative entropy generation rates and unbalanced component conservation. This provides accurate and reliable thermodynamic field quantity prediction support for subsequent entropy-minimization multi-objective optimization, while realizing seamless integration between the surrogate model and the core thermodynamic optimization objectives of this study.

To ensure the prediction accuracy and generalization performance of the thermodynamic-constrained PINN surrogate model, this study establishes a rigorous quality control system for simulation datasets and targeted training strategies. The core innovation lies in realizing coordinated optimization of data fidelity and training efficiency, addressing the shortcomings of traditional surrogate models such as uneven data quality and susceptibility to local optima during training. The CFD simulation dataset undergoes strict grid independence verification, with the total number of control volumes not less than 1.2 × 10⁶. Near-wall grids are refined to ensure y⁺ < 1, accurately capturing near-wall viscous sublayer flow and heat–mass transport characteristics, while strictly controlling the relative simulation error within 5%, thereby providing high-fidelity data support for model training and avoiding prediction deviations caused by low-quality data. The model training adopts a global iteration strategy of 5000 epochs, innovatively introducing an adaptive learning rate decay mechanism, with the initial learning rate set to 0.001 and reduced by 50% every 1000 iterations. This design ensures rapid convergence in the early stage while avoiding oscillations in the later stage, effectively balancing training convergence speed and global optimal solution search capability. The trained surrogate model can complete full-domain thermodynamic field quantity prediction for a single set of aesthetic parameters within 10 ms, achieving approximately a 10⁵-fold acceleration compared with traditional CFD simulations, greatly reducing the computational cost of subsequent multi-objective optimization, and providing technical assurance for efficient optimization of high-dimensional aesthetic parameters.

To enable precise quantification and comparison in multi-objective optimization, this study rigorously derives and defines two categories of core thermodynamic evaluation indicators based on non-equilibrium thermodynamic theory. The innovation lies in the fact that these indicators combine physical interpretability with comparability, addressing the defects of traditional evaluation indicators, which are ambiguously defined and unable to simultaneously consider full-domain dissipation and distribution uniformity. The global entropy generation number, as the core indicator for quantifying the degree of full-domain irreversible dissipation, is obtained by integrating the total entropy generation rate over the domain and normalizing it with respect to the thermal load. Its expression is:

$N_s=\frac{\iiint_V \dot{S}_{\text {gen}, \text { tot }} d V}{Q \cdot \Delta T}$           (8)

This indicator eliminates quantitative deviations under different thermal load conditions and directly reflects the influence of aesthetic parameter regulation on full-domain thermodynamic dissipation, providing a clear quantitative basis for the entropy-minimization objective. The exergy uniformity index is used to evaluate the spatial uniformity of indoor exergy distribution, expressed as:

$I_{e x}=1-\frac{\operatorname{std}\left(T_{e x}\right)}{\operatorname{mean}\left(T_{e x}\right)}, T_{e x}=T_0 \frac{e-e_0}{s-s_0}$          (9)

where, T₀, e₀, s₀ denote the reference state parameters of the standard environment, and T_ₑₓ represents the exergy temperature. By taking the ratio of the standard deviation to the mean, this indicator converts the uniformity of exergy distribution into a quantifiable numerical value, avoiding the limitation of traditional evaluations that focus only on average values while neglecting spatial distribution differences. These two indicators jointly constitute the core of thermodynamic evaluation, providing unified and precise quantitative criteria for subsequent multi-objective optimization, and highlighting the thermodynamics-led optimization philosophy of this study.

2.3 Multi-objective thermodynamic optimization algorithm

The core innovation of this study lies in constructing an entropy-dominated normalized multi-objective joint function, breaking through the limitation of imbalanced weighting among thermodynamic dissipation, exergy distribution, and architectural aesthetics in traditional multi-objective optimization, thereby realizing the collaborative optimization of the three and highlighting the core orientation of minimizing thermodynamic irreversible dissipation. This function integrates three categories of core indicators: full-domain irreversible dissipation, spatial uniformity of exergy, and objective architectural aesthetics. Normalization is applied to eliminate dimensional differences, ensuring that all objectives share a comparable basis during the optimization process. Its expression is:

$F=w_1 N_s+w_2\left(1-I_{e x}\right)-w_3 S_{m o r p h}$            (10)

The function is designed to minimize the objective function value. The first term characterizes the degree of full-domain thermodynamic irreversible dissipation, the second term reflects the non-uniformity of exergy distribution, and the third term quantifies the level of architectural form aesthetics. Weight allocation reinforces the dominant role of thermodynamic indicators. The weighting coefficients are strictly determined through the Analytic Hierarchy Process (AHP): the entropy generation-related weight w₁ takes a value of 0.5, the exergy uniformity weight w₂ takes a value of 0.3, and the architectural aesthetics weight w₃ takes a value of 0.2. This allocation balances exergy utilization efficiency and architectural aesthetics expression, avoiding performance imbalance caused by single-objective optimization.

To realize the objective quantification of architectural aesthetics and overcome the dependence of traditional aesthetic evaluation on subjective scoring, this study innovatively introduces morphological complexity entropy as the aesthetic quantification indicator, with the complete calculation formula expressed as:

$S_{m o r p h}=-\sum p_i \log p_i$           (11)

where, pi denotes the grayscale probability distribution within the indoor field of view, obtained through grayscale analysis of architectural form projection images. It precisely characterizes the orderliness, continuity, and complexity of architectural forms, realizing a non-subjective, quantitative aesthetic evaluation. The introduction of this indicator transforms architectural aesthetics from traditional qualitative description into a quantitative variable that can be incorporated into multi-objective optimization, forming an organic coupling with thermodynamic indicators. The entire multi-objective function not only ensures the core objective of minimizing thermodynamic dissipation but also achieves synergistic improvement of exergy uniformity and architectural aesthetics, resolving the core challenge in traditional design where aesthetics and low dissipation cannot be achieved simultaneously, while possessing clear physical meaning and engineering practicability. The flowchart of architectural visual aesthetics quantification is shown in Figure 3.

This study adopts the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) algorithm with elitist preservation strategy to conduct global optimization in high-dimensional aesthetic parameter space. The core innovation lies in introducing an original ventilation thermodynamic safety penalty term, addressing the core deficiency of traditional NSGA-II algorithms in building ventilation optimization, which tends to produce irrational design solutions and neglect thermodynamic safety boundaries. The algorithm flowchart is shown in Figure 4. The core hyperparameters of the algorithm are determined through multiple rounds of debugging and optimization: population size is set to 100, maximum generations to 150, crossover probability to 0.8, and mutation probability to 0.05. This parameter combination ensures global optimization capability while avoiding entrapment in local optima, balancing optimization efficiency and solution diversity. Compared with traditional optimization strategies, this study adds thermodynamic hard constraints, using the thermodynamic rationality of indoor ventilation as a prerequisite screening condition for optimization solutions, ensuring that the optimized results not only achieve optimality in the multi-objective function but also conform to the thermodynamic safety requirements of practical engineering applications.

Figure 3. Flowchart of architectural visual aesthetics quantification based on morphological complexity entropy

Figure 4. Flowchart of Non-Dominated Sorting Genetic Algorithm II (NSGA-II) multi-objective optimization incorporating a dual-layer verification mechanism and thermodynamic hard constraints

The core innovation is reflected in the quantitative construction and embedding of the ventilation thermodynamic safety penalty term, with its original quantitative rule expressed as:

$F^*=\left\{\begin{array}{c}F, R H<90 \%, \text { AirAge}<300 s \\ +\infty, \text { Other unreasonable conditions }\end{array}\right.$           (12)

This penalty term enforces the elimination of unreasonable design solutions through fitness reconstruction, with relative humidity and air age serving as key thermodynamic safety indicators that strictly define reasonable indoor ventilation conditions. When a design solution satisfies relative humidity below 90% and air age less than 300 seconds, its original objective function value is retained for evolutionary screening; when conditions such as high humidity or air stagnation occur, violating the thermodynamic rationality of indoor ventilation, the fitness is set to infinity, forcibly eliminating that design solution. This hard constraint design directly embeds thermodynamic safety boundaries into the optimization process, avoiding the drawback of traditional optimization that pursues only objective function optimality while neglecting actual feasibility, ensuring that the final optimized solutions possess thermodynamic superiority, aesthetic rationality, and engineering practicability, while reinforcing the thermodynamics-led optimization philosophy of this study.

To address the core limitation of traditional multi-objective optimization that relies solely on surrogate model prediction without physical reality verification, this study innovatively constructs a dual-layer verification mechanism, achieving dual guarantees of optimization efficiency and physical accuracy, thereby ensuring the reliability and engineering applicability of the final optimized solutions. The core of this mechanism lies in uniformly feeding the optimal candidate parameter combinations screened from the Pareto front back into a high-precision CFD model that has undergone grid independence verification for secondary numerical checking. The focus is on verifying the calculation deviations of the full-domain entropy generation rate, temperature field, and humidity field. By quantitatively comparing the surrogate model predictions with high-precision CFD simulation values, the deviation between the two is strictly controlled within 5%. This dual-layer verification design not only leverages the surrogate model for efficient optimization of high-dimensional parameters but also eliminates the propagation of surrogate model prediction errors through high-precision CFD secondary verification, preventing optimized solutions from deviating from actual thermodynamic responses due to surrogate model approximation, while strengthening the physical rigor of the optimization results.

The core academic contribution of this study lies in constructing an interdisciplinary technical system that deeply integrates non-equilibrium thermodynamics, architectural aesthetics, and intelligent optimization, realizing an essential transformation in indoor natural ventilation design from “indicator orientation” to “mechanism orientation,” and achieving theoretical breakthroughs and technological innovations in three core directions. The establishment of the aesthetics–thermodynamics coupling mechanism directly links architectural form aesthetic parameters with thermodynamic boundary conditions through analytical relationships for the first time, breaking the limitation of traditional research in which aesthetics and thermodynamics are separated. It elevates architectural aesthetics from a purely visual evaluation indicator to a core means capable of actively regulating irreversible dissipation, enriching the application scenarios of non-equilibrium thermodynamics in the field of architectural design, and filling the research gap in the interdisciplinary coupling of architectural aesthetics and thermodynamic dissipation. The innovation of the thermodynamic-constrained PINN surrogate model fundamentally solves the key problem of traditional data-driven surrogate models producing predictions detached from physical mechanisms, by embedding the incompressible Navier–Stokes equations and entropy generation conservation laws into the loss function. This greatly improves the computational efficiency and prediction accuracy of high-dimensional parameter optimization, and its technical concept can be extended to various fluid mechanics and thermodynamics coupled optimization scenarios, providing a new technical pathway for the efficient optimization of complex engineering systems. The entropy-dominated multi-objective evaluation system allocates weights centered on thermodynamic irreversible dissipation, realizing the collaborative optimization of low entropy generation, high aesthetics, and superior comfort, breaking through the limitations of traditional single-objective optimization and enhancing the academic competitiveness of the research.

An in-depth thermodynamic mechanism analysis of the experimental results indicates that architectural aesthetic parameters exert a clear physical law in regulating indoor ventilation irreversible dissipation, confirming the core hypothesis that “beauty corresponds to low dissipation.” When the window opening aspect ratio approaches the golden ratio constant, the degree of inlet flow distortion significantly decreases, turbulence vorticity dissipation weakens, and the proportion of viscous entropy generation declines. This is essentially because reasonable aesthetic proportions optimize the inlet flow field structure, reducing irreversible losses in the momentum transport process. Optimization of the air-guide tower taper angle and roof Gaussian curvature effectively regulates the balance between heat conduction entropy generation and viscous dissipation entropy generation by altering wall roughness and thermal boundary layer thickness, achieving precise control of the full-domain entropy generation rate. Experimental results under different climate scenarios further reveal that in hot-humid environments, humidity diffusion entropy generation is the main source of irreversible loss, and the optimization of aesthetic parameters should focus on reducing moisture diffusion resistance; whereas in hot-dry environments, it is necessary to balance heat conduction and viscous dissipation through optimization of curved surface morphology. This regularity provides a theoretical basis for climate-oriented aesthetics–thermodynamics collaborative optimization, enabling the optimization framework to adapt to the thermodynamic characteristics of different regions and enhancing its practical engineering application value.

Although this study has made significant progress in interdisciplinary integration and technological innovation, certain limitations remain. Currently, only five categories of core geometric indicators are selected as aesthetic parameters, without including material texture, color, and other aesthetic factors, whose influences on thermal radiation and heat conduction have not yet been fully considered. Numerical simulations adopt only steady-state sedentary office loads, without covering the disturbance effects of dynamic loads such as personnel activity and equipment heat dissipation on indoor thermodynamic fields. The verification methods rely mainly on numerical simulation and schlieren experiments, lacking long-term measured data from actual building prototypes, making it difficult to fully reflect the complex thermodynamic responses in real engineering scenarios.

Future research will focus on expanding the scope of aesthetic parameters, introducing material aesthetics and color thermodynamics to improve the comprehensiveness of the coupling model; embedding dynamic load prediction models to enhance the framework’s adaptability to actual working conditions; and constructing real building prototypes to verify the feasibility and stability of optimization schemes through long-term measurements. Compared with similar international studies, this research demonstrates significant innovation in thermodynamic constraint embedding and the aesthetics–thermodynamics direct coupling mechanism. Relative to similar studies published in International Journal of Heat and Mass Transferand Energy and Buildings, this work places greater emphasis on the deep integration of thermodynamic mechanisms and architectural aesthetics, and possesses evident advantages in both computational efficiency and optimization accuracy. Looking forward, the TAEM framework can be combined with digital twin and reinforcement learning technologies to further improve optimization efficiency and adaptive regulation capability, promoting the deep integration of non-equilibrium thermodynamics with architectural design and green low-carbon development, and providing more complete theoretical support and technical paradigms for the ventilation design of passive and zero-carbon buildings.